- A Survey on LLM Inference-Time Self-Improvement [15.0]
近年,テスト時の計算量の増加による推論向上技術が注目されている。 本稿では,最近の研究を包括的にレビューし,詳細な分類学に貢献し,課題と限界について議論する。
論文 参考訳（メタデータ） (Wed, 18 Dec 2024 21:37:07 GMT) - 最近要注目（？）のInference time self improvementのサーベイ。「We classify these methods into three categories: Independent Self-Improvement, which operates independently; Context-Aware Self-Improvement, which leverages external support (i.e. context and datastore retrieval); and Model-Aided Self-Improvement, which relies on external models for collaboration.」という軸で整理。
- リポジトリはGitHub – dongxiangjue/Awesome-LLM-Self-Improvement: A curated list of awesome LLM Inference-Time Self-Improvement (ITSI, pronounced “itsy”) papers from our recent survey: A Survey on Large Language Model Inference-Time Self-Improvement.